1. Introduction
The use of engineered nanomaterials (ENMs) in everyday products is constantly growing. Because of their nanoscale, these particles have fascinating properties in contrast to bulk material. For example, silicon dioxide is used in tires to reduce abrasion. Silver nanoparticles have antibacterial properties and are used in medical products and sportswear. Nanoscale cerium dioxide is used as a diesel fuel additive to reduce particulate matter emissions and fuel consumption in diesel engines. Nanoscale titanium dioxide is added to sunscreen because of its protective effect against UV radiation.
Nanoparticles are often used as additives and incorporated particularly in thermoplastics. There, they act, for example, as fillers, color pigments or flame retardants and enhance the mechanical, optical or thermal properties of thermoplastics strongly.
At the end of life (EoL), these products must also be safely and properly recycled or disposed of. This so-called “nanowaste” requires special attention as its processing, recycling and disposal may release nanoparticles, making it a possible hazard for humans and the environment [
1,
2].
Industrially produced nanoparticles with the highest production amounts are carbon black, titanium dioxide, and silicon dioxide (silica), which are generally produced by flame synthesis. However, there are no reliable figures regarding the production amounts of ENM in Germany, Europe or worldwide as there is no legal obligation to report them, with some exceptions in France, Belgium, Denmark, Norway and Sweden [
3]. To fill this gap, some authors [
4,
5,
6,
7,
8] have conducted surveys of experts in the industry, who in turn have estimated the production amounts. A comparison of these figures can be found in
Table 1, which also includes CeO
2 and carbon nanotubes (CNTs) and the mass-produced ENM.
The table shows that the individual estimates differ significantly, even though they took place at approximately the same time. The more recent the study, the higher the estimated production volumes, which correlates with the general growth of the nanomaterials market.
The same data gap exists concerning nano-enabled products. In general, there is very little information about which products contain nanomaterials, as there are only a few areas, such as cosmetics, food packaging and biocidal substances, where specific legislation exists in Germany and the EU [
10]. So far, only France, Denmark, Norway, Belgium, Italy and Sweden have registers in which products containing nanomaterials must be recorded [
11]. Within the category of metal oxide nanomaterials, titania is by far the most used nano additive [
12].
Titanium dioxide is mainly used as a white pigment in a wide range of products, and globally, several million tons of titanium dioxide are processed annually in plastics, coatings, paints, food, cosmetic products and pharmaceuticals [
13].
An admixture of nanoparticulate additives can have an impact on the entire process chain, from production to disposal. According to a German study by Conversio Market & Strategy GmbH [
13], about 142 kt of TiO
2 was incorporated into 14.4 million t of plastic products in Germany in 2017. Thus, on average, the TiO
2 content in plastics is about 1.0 wt.-%, with ⅔ of the TiO
2 amount used in plastic products in the packaging and construction sectors. Particularly high TiO
2 contents are present in white plastic profiles, such as window frames, with a TiO
2 content of 3 to 5%.
The unclear data situation regarding production quantities [
14] also makes it difficult to predict the possible amounts of nanomaterials released into the environment based on a life cycle analysis.
There are many potential pathways for the release of nanomaterials along their life cycle, whether during production, use, or recycling [
15,
16]. If recycling is possible, depending on the kind of product, it is always associated with downcycling. Ultimately, in Germany and various other countries, waste is thermally recovered. During this incineration, mineral waste content including ENM is to be converted into harmless ash for disposal or further use as construction material.
For this reason, we previously dealt with the behavior of nanomaterials during thermal utilization, investigated whether nanoparticles can be released in large-scale plants via the gas path, and determined the sinks within these plants for ceria particles [
17,
18,
19]. Cerium dioxide was used as tracer material because it is hardly found in the background of such plants. Therefore, the samples taken were analyzed for cerium quite easily. There are also studies where titanium dioxide has been introduced into incinerators, but since it occurs far more frequently, it is very difficult to distinguish the added titanium dioxide from that in the background, so the results of these studies are subject to significant fluctuations [
20].
In order to be able to predict the behavior of nanoparticles in the complex aerosol of combustion plants, fundamental studies are required. To our knowledge, very few studies have dealt with the addition of nanoparticles to flames. In one of these published studies, particles were dosed into flames in powder form and the effects on the particle size distribution were observed [
21]. SiO
2 was added to a methane/oxygen diffusion flame using a brush feeder. With an increasing equivalence ratio, a peak formed at significantly smaller particle sizes, and the distribution was bimodal until the peak disappeared completely at large diameters with further increasing equivalence ratio, where only the new peak could be seen. The processes were confirmed by means of particle dynamics modeling (General Dynamics Equation—GDE), whereby total or partial evaporation with subsequent nucleation, coagulation and surface growth was assumed. In another publication by the authors [
22], WO
3 particles were used with a modal value of the particle size distribution at approx. 700 nm. The temperatures used were also partly above the boiling temperature of WO
3, which is why the authors used purely physical processes, namely, evaporation, nucleation, coagulation and surface growth, to describe their experimental findings theoretically. As the distance to the burner increased, the number concentration in the new peak decreased and shifted slightly to the right. At a height of approx. 4 cm above the burner, the particle material completely passed into the gas phase, and only at a height of approx. 10 cm above the burner did homogeneous nucleation occur because of supersaturation. The predictions of the model (also GDE) could be confirmed experimentally.
At this point, it must also be briefly pointed out that very small nanoparticles exhibit different physical behavior than the respective bulk material. Buffat and Borel used gold particles to show that the melting temperature changes depending on the particle size [
23]. The 2.5 nm gold particles used had a melting temperature that is approx. 1000 °C lower than that of the bulk material. Other authors confirmed this finding [
24,
25,
26] and presented models for calculating these effects. Evaporation at temperatures well below the bulk evaporation temperature has also been observed and interpreted by some authors. In the physical process of NP vaporization, a linear relationship between the vaporization temperature and the reciprocal particle diameter has been reported [
24,
27,
28]. This relationship is also found in other size-dependent thermodynamic quantities, such as the melting temperature, melting enthalpy, melting entropy, Curie temperature and Debye temperature [
24]. Because of the significantly different properties of NP compared with the bulk material and the greatly altered thermodynamic behavior, a new interdisciplinary field, nanothermodynamics, has been established [
29]. However, the models only predict a strong deviation from the value of the bulk material for various thermodynamic variables below approx. 20 nm.
As described above, TiO2 is the most commonly used metal oxide nanomaterial, and studies on the thermal behavior of its particles are not yet available. Therefore, in this study, basic laboratory investigations on the influence of temperature on titania and ceria particles were performed to obtain comprehensible data sets. For this purpose, a well-defined quasi-adiabatic flame stabilized on a Bunsen-type burner was used. Titania and ceria nanoparticles are added to a laminar, premixed flame. With online particle sizing instruments as well as offline image analysis, the influence of flame temperature on the aerosol was investigated to explain and model the experienced phenomena. The results show that titania builds a new particle peak at very small particle sizes at high temperatures. Since the maximum flame temperatures are below the evaporation temperature of titania and the resulting particle size is smaller than the primary particles in the original titania agglomerates, enhanced vaporization of titania in the chemically reacting flame is considered a plausible explanation.
2. Materials and Methods
For basic studies at the laboratory scale, the focus was on titania material since it is more relevant to the production amounts than ceria. For the pilot and industrial-scale experiments, we used ceria since it has a very low background in such plants. Therefore, the ceria particles were studied at the laboratory scale as well, but they play a minor role in this paper.
As the flame, we used laminar ethylene–air flames diluted by argon in order to provide a stationary well-defined investigation system for the temperature impact on ENM.
This section is divided into 5 parts as follows: a laboratory Bunsen-type burner, the used nanoparticles, aerosol sampling and characterization, equilibrium temperature calculation and temperature measurement via coherent anti-Stokes Raman spectroscopy (CARS).
Figure 1 shows the experimental setup.
2.1. Laboratory Bunsen-Type Burner
For these investigations, we used a Bunsen-type burner. At the tube outlet, a conical, laminar, premixed flame was stabilized. Nanoparticle suspensions were premixed with the fuel/air mixture and added at the inlet of the burner.
The burner consists of a 70 cm long stainless-steel tube with an inner diameter of 10 mm. The gases were added at the inlet of the burner via 6 mm Swagelok connections. A cooling jacket encloses the gas-carrying tube, through which water flows at a fixed temperature so there is a constant gas inlet temperature at the burner exit.
Mass flow controllers (MFCs) (EL-Flow, Bronkhorst, Ruurlo, The Netherlands) set the volume flow of the needed gases. Synthetic air was used both as an oxidizer for the combustion of ethylene and as a carrier gas for the sprayed nanoparticle suspension. Argon was used to reduce the flame temperature while maintaining the stoichiometry of the combustion. The advantage of the burner design is that the nanoparticles and the fresh gas pass through the flame front together unhindered and no separation can occur, which would be the case with a flat flame burner, for example, where the flame is stabilized on a sintered metal plate. A list of the used settings regarding argon addition and volume flows can be found in
Table 2. The total volume flow was fixed to ensure the same cold gas velocity for all used settings. The equivalence ratio Φ is the ratio between the oxygen content in the gas mixture and that required for complete stoichiometric combustion, which was set to 1 for all argon admixtures.
2.2. Used Nanoparticles
The used nanoparticles are commercially available TiO2 (Aeroxide® P25, Evonik) and CeO2 (NanoArc CE-0440, Alfa Aesar—now available as NanoArc CE-6440, Thermo Fisher Scientific, Waltham, MA, USA), which are available in powder form and as a suspension, respectively.
Aeroxide
® P25 is available as a powder with crystalline TiO
2 with 85% anatase and 15% rutile modification. It is noteworthy that the more thermodynamically stable phase (rutile) is present to a much lesser extent despite the manufacturing conditions (flame hydrolysis). The rutile modification is concentrated on the particle surface and intensively interlocked with the anatase phase [
30].
The used CeO2 is commercially available as a suspension with 25 w.-% CeO2 and 75 w.-% water, produced by Alfa Aesar, which is now part of Thermo Fisher Scientific. It is produced via a gas-phase plasma process through which the precursor is evaporated. After quenching the vapor, CeO2 nuclei form and grow to the desired nanoparticle size. These nanoparticles are then transferred to a suspension.
For the investigations, a suspension with demineralized water was prepared at a concentration of 4 g/L. This suspension was treated in an ultrasonic bath for one hour before starting the experiment and continuously stirred during the experiment to avoid agglomeration and aging as far as possible. The suspension was sprayed into the air by means of an atomizer (ATM220, Topas GmbH, Dresden, Germany) and added together with the premixed fuel gas at the inlet of the burner. The dosed mass flow of the suspension was determined at each test and was approximately 1.5 g/h.
2.3. Aerosol Sampling and Characterization
Above the burner, there is a tightly closing protective cylinder (length: 505 mm, inner diameter: 96 mm) made of quartz glass, which, on the one hand, prevents false air from leaking in and, on the other hand, offers a connection piece at different heights above the burner (HAB) for the connection of a sampling probe. In this study, the sampling probe, which is a 90° curved steel probe with a length of 172.5 mm and an inner diameter of 6 mm, was always installed in a HAB of approximately 450 mm. This corresponds to the highest possible measurement position. A dilution stage (VKL10E, Palas GmbH, Karlsruhe, Germany) upstream of the sampling probe diluted the number concentration of the aerosol to slow down coagulation processes. Additionally, it cooled down the aerosol sufficiently to be collected by the measuring device. The dilution stage has an input volume flow of 1.88 L/min (standard volume flow rate at 273.15 K and 1.01325 bar) and a dilution air volume flow of 20 L/min, resulting in a dilution factor of 11.6. The input flow rate was checked by a flow calibrator (Flow Calibrator 4148, TSI GmbH, Aachen, Germany), and the dilution air flow rate was set by a mass flow controller (EL-Flow, Bronkhorst, Ruurlo, The Netherlands). The particle size distribution was measured using a scanning mobility particle sizer (SMPS+C, Grimm, Ainring, Germany), which was operated with the M-DMA (Differential Mobility Analyzer, size range 5 to 350 nm), allowing mobility-equivalent diameters to be measured. The calibration of the mobility analyzer was performed with particle standards. Latex particles were used here because they are easy to handle and have a known density and spherical shape.
As a further measurement technique, the Electrical Low-Pressure Impactor (ELPI+, Dekati Ltd., Kangasala, Finland) was used. The ELPI consists of 15 impactor stages in a cascade setup. Before entering the cascade impactor, the particles are charged by passing a cylindrical tube corona charger. The charger efficiency is size-dependent and well-defined [
31]. The mean aerodynamic diameter, which is the classifying physical quantity, ranges between 6 nm and 9.9 µm. The last stage (D50 = 6 nm) is a backup filter and collects all particles remaining in the gas flow. Each impactor stage is connected to an electrometer to measure the current of the impacted particles and to transform it into a number size distribution. It is possible to place foils or grids on each impactor stage to analyze the deposited particles via imaging techniques afterward.
For imaging, a transmission electron microscope TEM (EM910 Leo, Carl Zeiss Microscopy GmbH, Oberkochen, Germany) was used. The TEM grids were sampled in different ways depending on the research question. (a) To sample the original agglomerates without passing the flame, a TEM grid was installed in a filter housing at HAB ≈ 45 cm. (b) The sampling technique for imaging of aerosol particles that went through the flame included installing the TEM grid on the impaction plate of a low-pressure impactor where the complete aerosol is impacted. (c) For the comparison of volume equivalent diameter and aerodynamic diameter, TEM grids were installed on each ELPI impaction stage, where each stage corresponds to one specific aerodynamic diameter.
2.4. Equilibrium Temperature Calculations
For the comparison of the measured temperatures via CARS, equilibrium calculations were executed using ANSYS Chemkin [
32] to provide values for the adiabatic flame temperatures with different flame stoichiometries and argon fractions. With the “equilibrium reactor”, it is possible to perform chemical and phase equilibrium calculations. GRImech 3.0 was used as a chemistry set for preprocessing, which includes thermodynamic data, transport data and gas-phase reactions for hydrocarbon combustion up to C
3. The inlet gas temperature was set to 298 K, the pressure to 1 atm, and the problem type to “Constant Pressure Enthalpy”. Ethylene (C
2H
4) acted as the fuel and synthetic air (79% N
2, 21% O
2) as the oxidizer.
2.5. Temperature Measurement via CARS
As a non-intrusive laser diagnostic, coherent anti-Stokes Raman spectroscopy (CARS) has been well established as a reliable quantitative thermometry in various combustion environments [
33]. CARS is nonlinear spectroscopy that provides spatially and temporally resolved temperature and species concentrations by probing molecular Raman shifts. Three coherent laser beams (pump, Stokes and probe) are focused and crossed in the region of interest, generating a CARS signal beam. The wavelengths of the three beams are chosen such that their interactions excite specific molecular ro-vibrational transitions of the target molecule. A broadband Stokes beam in combination with narrowband pump and probe beams (usually from the same laser source) is commonly employed to cover a wide range of vibrational and rotational transitions simultaneously. The N
2 Q-branch is often adopted because of the abundance of N
2 in the combustion environment. The resulting anti-Stokes signal beam carries the Raman spectra of N
2. By comparing the experimentally obtained spectra to theoretical ones, the ro-vibrational temperature of N
2 can be derived, which is assumed to be in equilibrium with the gaseous kinetic temperature in the measurement volume.
In this study, a frequency-doubled Nd:YAG laser (Quanta Ray Pro 290, Spectra Physics, Stahnsdorf, Germany) with a repetition rate of 10 Hz (λ = 532 nm, pulse duration ∼8 ns) was used to pump two dye lasers (Sirah Double Dye PrecisonScan, Sirah Lasertechnik, Grevenbroich, Germany). A narrowband dye laser was tuned to produce 591 nm, which provided the pump and probe beams (50% split). Furthermore, the broadband dye laser (Stokes beam) was set to peak around 685 nm with a bandwidth of about 10 nm. Compared with the often-used wavelengths of 532 nm and 607 nm, the choices of these specific wavelengths shift the CARS signal from 473 nm to 519 nm, hence avoiding signal interference due to laser-induced C
2 emissions, a problem often encountered in sooting flames [
34]. The lasers and the related optical components were mounted in a mobile container as detailed in [
35]. To obtain a high-intensity CARS signal while maintaining high spatial resolution, a folded BOXCARS configuration [
34] was used to achieve the necessary phase matching required for generating coherent CARS signal beams. Outside the laser container, the three laser beams (pump, Stokes and probe) were relayed by a series of high-reflectivity mirrors to the measurement position. An achromatic focal lens (f = 250 mm) was used to focus the three beams on the probe volume in the center plane of the burner. Prior to each measurement, a removable beam splitter was placed in front of the measurement position to deflect about 2% of the beams to a beam profiling camera (WinCamD, DataRay, Redding, CA, USA). The camera was placed at the focal plane of the three beams and was used to assist in fine adjustments of the beam overlaps. Spatial resolution was determined to be <0.1 mm in diameter (using the beam profiling camera) and about 1.5 mm (L95%) in the laser beam propagating direction using the non-resonant CARS signal from a thin quartz plate placed at the measurement position. At the crossing point of the three laser beams, the 519 nm CARS anti-Stokes signal beam was generated in room air or flame environments, which traveled coherently in the same direction as the other laser beams. It was captured by a plano-convex lens (f = 350) and directed into a spectrometer (THR 1000, Jobin-Yvon ISA Instruments, Longjumeau, France) via an optical fiber and a series of focal lenses. The spectrometer was equipped with an 1800 lines/mm diffraction grating and an entrance slit set to 50 μm. A bandpass filter centered at 520 nm (ET520/20 m, Chroma, Bellows Falls, VT, USA) was placed in front of the entrance slit to eliminate signal interferences from pump lasers. The CARS signal was imaged onto an intensified charge-coupled device (ICCD) camera (PI-MAX Gen III, 1340 × 700 pixels, Princeton Instruments, Krailing, Germany). The images were binned on-chip to 335 superpixels along the spectral axis and 1 superpixel in the vertical direction (i.e., 2D image converted into a 1D spectrum) to enhance the signal-to-noise ratio. For this investigation, each spectrum was obtained with an on-CCD accumulation of 300 single shots to further improve the signal level. Single-shot measurements were also carried out at selected positions for comparison and to characterize local fluctuations. The optics were mounted on translation stages so that different axial and radial locations in the flame could be probed.
Premixed C
2H
4–air–argon flames at various equivalence ratios and argon dilutions were measured in this study. All stable operating conditions examined are listed in
Table 2. A radial symmetry was assumed for the stable flame, but asymmetric behavior was observed for certain conditions deviating from the reference flame (0% argon, Φ = 1). For cases other than the reference flame, only x positions were varied with r fixed at 0.
Prior to fitting the experimentally obtained spectra, the pixel axis needed to be converted to a wavelength (spectral) axis. For this, emission from a mercury–argon lamp (Avalight-Cal-Ar, Avantes, Apeldoorn, The Netherlands) was coupled into the spectrometer using the same fiber optic used for the CARS signal. In addition, a tungsten lamp was used to create a flat background profile to correct for any wavelength-dependent quantum yield (signal intensity) of the spectrometer as well as the camera.
Each CARS spectrum was accompanied by a background spectrum that was acquired right after the CARS spectrum and by blocking the laser beams. During the measurement campaign, pure argon flow was used regularly to obtain the spectral profile of the broadband Stokes beam (based on the non-resonant background), which directly affects the spectral intensity distribution of the CARS signal. In the end, each CARS spectrum prior to a fitting procedure was (1) subtracted by the background noise, (2) corrected for nonuniformities in the Stokes profile and (3) converted to a wavelength (wavenumber) axis.
For spectral fitting, a theoretical CARS spectrum was first generated and then convoluted with the laser profile (assumed Gaussian) and the slit function of the spectrometer. The slit function of the spectrometer was obtained by fitting a CARS spectrum taken in room air with known temperature with an asymmetric Voigt function [
36]. The pump/probe laser linewidth was assumed to be 0.2 cm
−1, as specified by the manufacturer. The exact value of the laser linewidth was found to exert negligible influence on the fitted temperature. Experimental spectra were then fitted iteratively against theoretical ones using a CARS-fitting routine “CARSpy” [
37]. The routine was constructed to incorporate various well-known processes in the CARS process, including (but not limited to) the cross-coherence convolution of CARS susceptibility with pump/probe laser profiles, collision narrowing at moderate to high number density conditions (i.e., low temperature and/or high pressure) and Doppler broadening at high-temperature conditions [
36,
38]. Additionally, chemical equilibrium was assumed to calculate the local composition (for non-resonant background estimation) using the fitted temperature, and this was also included in the iterative fitting process. Since the mechanism used (GRI 3.0) is unreliable at low temperatures, the initial mixture composition was assumed to be intact until T = 1200 K. For the least-square fit, temperature was the only variable parameter in the theoretic model to minimize the fitting dependencies on inter-parameter correlations (e.g., among temperature, N
2 mole fraction and non-resonant background).